Localisation of Speech Technology is a massive task.
For command-and-control ASR, it is mainly the very time-consuming task of collecting and annotating a lot of data.
But a good quality TTS system requires deep knowledge of the morphology, syntax and phonology of the language
as well as the technical skill to build the TTS system.
|
|
Because of this, it is the process of building TTS systems
in new languages that has been the initial focus of LLSTI. Many of the technologies required in a TTS system are
also useful for other speech and language-related applications such as Dialogue Systems, Machine Translation
and Information Retrieval.
|
|
This is the methodology which LLSTI has proven for the building of a new language:
|
|
1. Provide a toolset for linguists & software engineers who are interested in
creating a TTS system in their local language.
2. Recruit participants from the relevant geographical area
3. Arrange training of participants in the tools and techniques needed
4. As funds allow, source equipment to do the data collection and transcription work
5. Support the participants whilst they do the work
6. Compile the resulting TTS systems into a publicly-available open-source code
repository
|
|
The toolset is based around
the open-source tools Festival and Festvox.
|
|
Top
|